Instructions to use Open4bits/DeepSeek-R1-mlx-2Bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Open4bits/DeepSeek-R1-mlx-2Bit with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Open4bits/DeepSeek-R1-mlx-2Bit", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Open4bits/DeepSeek-R1-mlx-2Bit", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("Open4bits/DeepSeek-R1-mlx-2Bit", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

MLX

How to use Open4bits/DeepSeek-R1-mlx-2Bit with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("Open4bits/DeepSeek-R1-mlx-2Bit")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

vLLM

How to use Open4bits/DeepSeek-R1-mlx-2Bit with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Open4bits/DeepSeek-R1-mlx-2Bit"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Open4bits/DeepSeek-R1-mlx-2Bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Open4bits/DeepSeek-R1-mlx-2Bit

SGLang

How to use Open4bits/DeepSeek-R1-mlx-2Bit with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Open4bits/DeepSeek-R1-mlx-2Bit" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Open4bits/DeepSeek-R1-mlx-2Bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Open4bits/DeepSeek-R1-mlx-2Bit" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Open4bits/DeepSeek-R1-mlx-2Bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

MLX LM

How to use Open4bits/DeepSeek-R1-mlx-2Bit with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "Open4bits/DeepSeek-R1-mlx-2Bit"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "Open4bits/DeepSeek-R1-mlx-2Bit"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "Open4bits/DeepSeek-R1-mlx-2Bit",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Docker Model Runner
How to use Open4bits/DeepSeek-R1-mlx-2Bit with Docker Model Runner:
```
docker model run hf.co/Open4bits/DeepSeek-R1-mlx-2Bit
```

DeepSeek-R1-mlx-2Bit

File size: 2,383 Bytes

b60d894
 
 
 
 
50a910d
b60d894
50a910d

---
license: mit
library_name: transformers
tags:
- mlx
- open4bits
base_model: deepseek-ai/DeepSeek-R1
pipeline_tag: text-generation
---

# Open4bits / DeepSeek-R1-MLX-2Bit

This repository provides the **DeepSeek-R1 model quantized to 2-bit in MLX format**, published by Open4bits to enable highly efficient local inference with minimal memory usage and broad hardware compatibility.

The underlying DeepSeek-R1 model and architecture are **developed and owned by DeepSeek AI**. This repository contains only a 2-bit quantized MLX conversion of the original model weights.

The model is designed for lightweight, high-performance text generation and instruction-following tasks, making it well suited for resource-constrained and local deployments.

---

## Model Overview

DeepSeek-R1 is a transformer-based large language model developed for strong general language understanding and generation.
This release provides a **2-bit quantized checkpoint in MLX format**, enabling efficient inference on CPUs and supported accelerators with reduced memory footprint.

Open4bits has started supporting **MLX models** to broaden compatibility with emerging quantization formats and efficient runtimes.

---

## Model Details

* **Base Model:** DeepSeek-R1
* **Quantization:** 2-bit
* **Format:** MLX
* **Task:** Text generation, instruction following
* **Weight tying:** Preserved
* **Compatibility:** MLX-enabled inference engines and efficient runtimes

This quantized release is designed to balance strong generation performance with low resource requirements.

---

## Intended Use

This model is intended for:

* Local text generation and conversational applications
* CPU-based or low-resource deployments
* Research, prototyping, and experimentation
* Self-hosted or offline AI systems

---

## Limitations

* Reduced performance compared to full-precision variants
* Output quality depends on prompt design and inference settings
* Not specifically tuned for highly specialized or domain-specific tasks

---

## License

This model follows the **MIT** as defined by the base model creators.
Users must comply with the licensing conditions of the base DeepSeek-R1 model.

---

## Support

If you find this model useful, please consider supporting the project.
Your support helps Open4bits continue releasing and maintaining high-quality, efficient open models for the community.